(Yicai Global) June 2 -- Less than two months after releasing large language model Tongyi Qianwen, Alibaba Group Holding's cloud computing arm has begun beta testing Tongyi Tingwu, its audio- and video-focused artificial intelligence model.

Tongyi Tingwu can complete the transcription, retrieval, summarization, and sorting of audio and video content in real-time, according to the demonstration of its capabilities. As an LLM, it can automatically take notes, organize interviews, and extract slides.

The predecessor of Tongyi Tingwu was an internal function of Alibaba Cloud called Tingwu, which mainly provided audio transcription for project meetings of its investment, human resources, and other departments, Yan Zhijie, Tingwu's head of technology, told Yicai Global. Starting last year, the firm began to mine the tool’s greater potential, Yan added.

Tongyi Tingwu can record in real-time, translate languages, summarize speeches, and review questions, said Zhou Jingren, Alibaba Cloud's Chief Technology Officer. In terms of video experience, Tongyi Tingwu supports chapter overview, imports and uploads from cloud and local disks, and summarizes full texts.

In response to concerns about the data security and privacy with LLMs, Zhou said Tongyi Qianwen’s training does not rely on user data, but more on basic knowledge. “Today's Tongyi Tingwu is born from the combination of basic knowledge with meeting and video scenarios, which is more about using the ability of technical models to help us do summaries, translations, and content extraction,” Zhou noted.

On April 11, Alibaba Cloud announced that it will integrate its LLM Tongyi Qianwen into all its apps and give corporate users access. “We keep reducing the cost and sources of using Tongyi Qianwen through a series of optimizations in the bottom layer of the cloud to serve the community at a large scale,” Zhou pointed out.

Editor: Futura Costaglione